4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
8 #include <linux/init.h>
10 #include <linux/fcntl.h>
11 #include <linux/slab.h>
12 #include <linux/kmod.h>
13 #include <linux/major.h>
14 #include <linux/device_cgroup.h>
15 #include <linux/highmem.h>
16 #include <linux/blkdev.h>
17 #include <linux/module.h>
18 #include <linux/blkpg.h>
19 #include <linux/magic.h>
20 #include <linux/buffer_head.h>
21 #include <linux/swap.h>
22 #include <linux/pagevec.h>
23 #include <linux/writeback.h>
24 #include <linux/mpage.h>
25 #include <linux/mount.h>
26 #include <linux/uio.h>
27 #include <linux/namei.h>
28 #include <linux/log2.h>
29 #include <linux/cleancache.h>
30 #include <linux/aio.h>
31 #include <asm/uaccess.h>
35 struct block_device bdev;
36 struct inode vfs_inode;
39 static const struct address_space_operations def_blk_aops;
41 static inline struct bdev_inode *BDEV_I(struct inode *inode)
43 return container_of(inode, struct bdev_inode, vfs_inode);
46 inline struct block_device *I_BDEV(struct inode *inode)
48 return &BDEV_I(inode)->bdev;
50 EXPORT_SYMBOL(I_BDEV);
53 * Move the inode from its current bdi to a new bdi. If the inode is dirty we
54 * need to move it onto the dirty list of @dst so that the inode is always on
57 static void bdev_inode_switch_bdi(struct inode *inode,
58 struct backing_dev_info *dst)
60 struct backing_dev_info *old = inode->i_data.backing_dev_info;
62 if (unlikely(dst == old)) /* deadlock avoidance */
64 bdi_lock_two(&old->wb, &dst->wb);
65 spin_lock(&inode->i_lock);
66 inode->i_data.backing_dev_info = dst;
67 if (inode->i_state & I_DIRTY)
68 list_move(&inode->i_wb_list, &dst->wb.b_dirty);
69 spin_unlock(&inode->i_lock);
70 spin_unlock(&old->wb.list_lock);
71 spin_unlock(&dst->wb.list_lock);
74 /* Kill _all_ buffers and pagecache , dirty or not.. */
75 void kill_bdev(struct block_device *bdev)
77 struct address_space *mapping = bdev->bd_inode->i_mapping;
79 if (mapping->nrpages == 0)
83 truncate_inode_pages(mapping, 0);
85 EXPORT_SYMBOL(kill_bdev);
87 /* Invalidate clean unused buffers and pagecache. */
88 void invalidate_bdev(struct block_device *bdev)
90 struct address_space *mapping = bdev->bd_inode->i_mapping;
92 if (mapping->nrpages == 0)
96 lru_add_drain_all(); /* make sure all lru add caches are flushed */
97 invalidate_mapping_pages(mapping, 0, -1);
98 /* 99% of the time, we don't need to flush the cleancache on the bdev.
99 * But, for the strange corners, lets be cautious
101 cleancache_invalidate_inode(mapping);
103 EXPORT_SYMBOL(invalidate_bdev);
105 int set_blocksize(struct block_device *bdev, int size)
107 /* Size must be a power of two, and between 512 and PAGE_SIZE */
108 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
111 /* Size cannot be smaller than the size supported by the device */
112 if (size < bdev_logical_block_size(bdev))
115 /* Don't change the size if it is same as current */
116 if (bdev->bd_block_size != size) {
118 bdev->bd_block_size = size;
119 bdev->bd_inode->i_blkbits = blksize_bits(size);
125 EXPORT_SYMBOL(set_blocksize);
127 int sb_set_blocksize(struct super_block *sb, int size)
129 if (set_blocksize(sb->s_bdev, size))
131 /* If we get here, we know size is power of two
132 * and it's value is between 512 and PAGE_SIZE */
133 sb->s_blocksize = size;
134 sb->s_blocksize_bits = blksize_bits(size);
135 return sb->s_blocksize;
138 EXPORT_SYMBOL(sb_set_blocksize);
140 int sb_min_blocksize(struct super_block *sb, int size)
142 int minsize = bdev_logical_block_size(sb->s_bdev);
145 return sb_set_blocksize(sb, size);
148 EXPORT_SYMBOL(sb_min_blocksize);
151 blkdev_get_block(struct inode *inode, sector_t iblock,
152 struct buffer_head *bh, int create)
154 bh->b_bdev = I_BDEV(inode);
155 bh->b_blocknr = iblock;
156 set_buffer_mapped(bh);
161 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
162 loff_t offset, unsigned long nr_segs)
164 struct file *file = iocb->ki_filp;
165 struct inode *inode = file->f_mapping->host;
167 return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
168 nr_segs, blkdev_get_block, NULL, NULL, 0);
171 int __sync_blockdev(struct block_device *bdev, int wait)
176 return filemap_flush(bdev->bd_inode->i_mapping);
177 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
181 * Write out and wait upon all the dirty data associated with a block
182 * device via its mapping. Does not take the superblock lock.
184 int sync_blockdev(struct block_device *bdev)
186 return __sync_blockdev(bdev, 1);
188 EXPORT_SYMBOL(sync_blockdev);
191 * Write out and wait upon all dirty data associated with this
192 * device. Filesystem data as well as the underlying block
193 * device. Takes the superblock lock.
195 int fsync_bdev(struct block_device *bdev)
197 struct super_block *sb = get_super(bdev);
199 int res = sync_filesystem(sb);
203 return sync_blockdev(bdev);
205 EXPORT_SYMBOL(fsync_bdev);
208 * freeze_bdev -- lock a filesystem and force it into a consistent state
209 * @bdev: blockdevice to lock
211 * If a superblock is found on this device, we take the s_umount semaphore
212 * on it to make sure nobody unmounts until the snapshot creation is done.
213 * The reference counter (bd_fsfreeze_count) guarantees that only the last
214 * unfreeze process can unfreeze the frozen filesystem actually when multiple
215 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
216 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
219 struct super_block *freeze_bdev(struct block_device *bdev)
221 struct super_block *sb;
224 mutex_lock(&bdev->bd_fsfreeze_mutex);
225 if (++bdev->bd_fsfreeze_count > 1) {
227 * We don't even need to grab a reference - the first call
228 * to freeze_bdev grab an active reference and only the last
229 * thaw_bdev drops it.
231 sb = get_super(bdev);
233 mutex_unlock(&bdev->bd_fsfreeze_mutex);
237 sb = get_active_super(bdev);
240 error = freeze_super(sb);
242 deactivate_super(sb);
243 bdev->bd_fsfreeze_count--;
244 mutex_unlock(&bdev->bd_fsfreeze_mutex);
245 return ERR_PTR(error);
247 deactivate_super(sb);
250 mutex_unlock(&bdev->bd_fsfreeze_mutex);
251 return sb; /* thaw_bdev releases s->s_umount */
253 EXPORT_SYMBOL(freeze_bdev);
256 * thaw_bdev -- unlock filesystem
257 * @bdev: blockdevice to unlock
258 * @sb: associated superblock
260 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
262 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
266 mutex_lock(&bdev->bd_fsfreeze_mutex);
267 if (!bdev->bd_fsfreeze_count)
271 if (--bdev->bd_fsfreeze_count > 0)
277 error = thaw_super(sb);
279 bdev->bd_fsfreeze_count++;
280 mutex_unlock(&bdev->bd_fsfreeze_mutex);
284 mutex_unlock(&bdev->bd_fsfreeze_mutex);
287 EXPORT_SYMBOL(thaw_bdev);
289 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
291 return block_write_full_page(page, blkdev_get_block, wbc);
294 static int blkdev_readpage(struct file * file, struct page * page)
296 return block_read_full_page(page, blkdev_get_block);
299 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
300 loff_t pos, unsigned len, unsigned flags,
301 struct page **pagep, void **fsdata)
303 return block_write_begin(mapping, pos, len, flags, pagep,
307 static int blkdev_write_end(struct file *file, struct address_space *mapping,
308 loff_t pos, unsigned len, unsigned copied,
309 struct page *page, void *fsdata)
312 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
315 page_cache_release(page);
322 * for a block special file file_inode(file)->i_size is zero
323 * so we compute the size by hand (just as in block_read/write above)
325 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
327 struct inode *bd_inode = file->f_mapping->host;
331 mutex_lock(&bd_inode->i_mutex);
332 size = i_size_read(bd_inode);
340 offset += file->f_pos;
346 if (offset >= 0 && offset <= size) {
347 if (offset != file->f_pos) {
348 file->f_pos = offset;
353 mutex_unlock(&bd_inode->i_mutex);
357 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
359 struct inode *bd_inode = filp->f_mapping->host;
360 struct block_device *bdev = I_BDEV(bd_inode);
363 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
368 * There is no need to serialise calls to blkdev_issue_flush with
369 * i_mutex and doing so causes performance issues with concurrent
370 * O_SYNC writers to a block device.
372 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
373 if (error == -EOPNOTSUPP)
378 EXPORT_SYMBOL(blkdev_fsync);
384 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
385 static struct kmem_cache * bdev_cachep __read_mostly;
387 static struct inode *bdev_alloc_inode(struct super_block *sb)
389 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
392 return &ei->vfs_inode;
395 static void bdev_i_callback(struct rcu_head *head)
397 struct inode *inode = container_of(head, struct inode, i_rcu);
398 struct bdev_inode *bdi = BDEV_I(inode);
400 kmem_cache_free(bdev_cachep, bdi);
403 static void bdev_destroy_inode(struct inode *inode)
405 call_rcu(&inode->i_rcu, bdev_i_callback);
408 static void init_once(void *foo)
410 struct bdev_inode *ei = (struct bdev_inode *) foo;
411 struct block_device *bdev = &ei->bdev;
413 memset(bdev, 0, sizeof(*bdev));
414 mutex_init(&bdev->bd_mutex);
415 INIT_LIST_HEAD(&bdev->bd_inodes);
416 INIT_LIST_HEAD(&bdev->bd_list);
418 INIT_LIST_HEAD(&bdev->bd_holder_disks);
420 inode_init_once(&ei->vfs_inode);
421 /* Initialize mutex for freeze. */
422 mutex_init(&bdev->bd_fsfreeze_mutex);
425 static inline void __bd_forget(struct inode *inode)
427 list_del_init(&inode->i_devices);
428 inode->i_bdev = NULL;
429 inode->i_mapping = &inode->i_data;
432 static void bdev_evict_inode(struct inode *inode)
434 struct block_device *bdev = &BDEV_I(inode)->bdev;
436 truncate_inode_pages(&inode->i_data, 0);
437 invalidate_inode_buffers(inode); /* is it needed here? */
439 spin_lock(&bdev_lock);
440 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
441 __bd_forget(list_entry(p, struct inode, i_devices));
443 list_del_init(&bdev->bd_list);
444 spin_unlock(&bdev_lock);
447 static const struct super_operations bdev_sops = {
448 .statfs = simple_statfs,
449 .alloc_inode = bdev_alloc_inode,
450 .destroy_inode = bdev_destroy_inode,
451 .drop_inode = generic_delete_inode,
452 .evict_inode = bdev_evict_inode,
455 static struct dentry *bd_mount(struct file_system_type *fs_type,
456 int flags, const char *dev_name, void *data)
458 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
461 static struct file_system_type bd_type = {
464 .kill_sb = kill_anon_super,
467 static struct super_block *blockdev_superblock __read_mostly;
469 void __init bdev_cache_init(void)
472 static struct vfsmount *bd_mnt;
474 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
475 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
476 SLAB_MEM_SPREAD|SLAB_PANIC),
478 err = register_filesystem(&bd_type);
480 panic("Cannot register bdev pseudo-fs");
481 bd_mnt = kern_mount(&bd_type);
483 panic("Cannot create bdev pseudo-fs");
484 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
488 * Most likely _very_ bad one - but then it's hardly critical for small
489 * /dev and can be fixed when somebody will need really large one.
490 * Keep in mind that it will be fed through icache hash function too.
492 static inline unsigned long hash(dev_t dev)
494 return MAJOR(dev)+MINOR(dev);
497 static int bdev_test(struct inode *inode, void *data)
499 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
502 static int bdev_set(struct inode *inode, void *data)
504 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
508 static LIST_HEAD(all_bdevs);
510 struct block_device *bdget(dev_t dev)
512 struct block_device *bdev;
515 inode = iget5_locked(blockdev_superblock, hash(dev),
516 bdev_test, bdev_set, &dev);
521 bdev = &BDEV_I(inode)->bdev;
523 if (inode->i_state & I_NEW) {
524 bdev->bd_contains = NULL;
525 bdev->bd_super = NULL;
526 bdev->bd_inode = inode;
527 bdev->bd_block_size = (1 << inode->i_blkbits);
528 bdev->bd_part_count = 0;
529 bdev->bd_invalidated = 0;
530 inode->i_mode = S_IFBLK;
532 inode->i_bdev = bdev;
533 inode->i_data.a_ops = &def_blk_aops;
534 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
535 inode->i_data.backing_dev_info = &default_backing_dev_info;
536 spin_lock(&bdev_lock);
537 list_add(&bdev->bd_list, &all_bdevs);
538 spin_unlock(&bdev_lock);
539 unlock_new_inode(inode);
544 EXPORT_SYMBOL(bdget);
547 * bdgrab -- Grab a reference to an already referenced block device
548 * @bdev: Block device to grab a reference to.
550 struct block_device *bdgrab(struct block_device *bdev)
552 ihold(bdev->bd_inode);
555 EXPORT_SYMBOL(bdgrab);
557 long nr_blockdev_pages(void)
559 struct block_device *bdev;
561 spin_lock(&bdev_lock);
562 list_for_each_entry(bdev, &all_bdevs, bd_list) {
563 ret += bdev->bd_inode->i_mapping->nrpages;
565 spin_unlock(&bdev_lock);
569 void bdput(struct block_device *bdev)
571 iput(bdev->bd_inode);
574 EXPORT_SYMBOL(bdput);
576 static struct block_device *bd_acquire(struct inode *inode)
578 struct block_device *bdev;
580 spin_lock(&bdev_lock);
581 bdev = inode->i_bdev;
583 ihold(bdev->bd_inode);
584 spin_unlock(&bdev_lock);
587 spin_unlock(&bdev_lock);
589 bdev = bdget(inode->i_rdev);
591 spin_lock(&bdev_lock);
592 if (!inode->i_bdev) {
594 * We take an additional reference to bd_inode,
595 * and it's released in clear_inode() of inode.
596 * So, we can access it via ->i_mapping always
599 ihold(bdev->bd_inode);
600 inode->i_bdev = bdev;
601 inode->i_mapping = bdev->bd_inode->i_mapping;
602 list_add(&inode->i_devices, &bdev->bd_inodes);
604 spin_unlock(&bdev_lock);
609 static inline int sb_is_blkdev_sb(struct super_block *sb)
611 return sb == blockdev_superblock;
614 /* Call when you free inode */
616 void bd_forget(struct inode *inode)
618 struct block_device *bdev = NULL;
620 spin_lock(&bdev_lock);
621 if (!sb_is_blkdev_sb(inode->i_sb))
622 bdev = inode->i_bdev;
624 spin_unlock(&bdev_lock);
627 iput(bdev->bd_inode);
631 * bd_may_claim - test whether a block device can be claimed
632 * @bdev: block device of interest
633 * @whole: whole block device containing @bdev, may equal @bdev
634 * @holder: holder trying to claim @bdev
636 * Test whether @bdev can be claimed by @holder.
639 * spin_lock(&bdev_lock).
642 * %true if @bdev can be claimed, %false otherwise.
644 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
647 if (bdev->bd_holder == holder)
648 return true; /* already a holder */
649 else if (bdev->bd_holder != NULL)
650 return false; /* held by someone else */
651 else if (bdev->bd_contains == bdev)
652 return true; /* is a whole device which isn't held */
654 else if (whole->bd_holder == bd_may_claim)
655 return true; /* is a partition of a device that is being partitioned */
656 else if (whole->bd_holder != NULL)
657 return false; /* is a partition of a held device */
659 return true; /* is a partition of an un-held device */
663 * bd_prepare_to_claim - prepare to claim a block device
664 * @bdev: block device of interest
665 * @whole: the whole device containing @bdev, may equal @bdev
666 * @holder: holder trying to claim @bdev
668 * Prepare to claim @bdev. This function fails if @bdev is already
669 * claimed by another holder and waits if another claiming is in
670 * progress. This function doesn't actually claim. On successful
671 * return, the caller has ownership of bd_claiming and bd_holder[s].
674 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
678 * 0 if @bdev can be claimed, -EBUSY otherwise.
680 static int bd_prepare_to_claim(struct block_device *bdev,
681 struct block_device *whole, void *holder)
684 /* if someone else claimed, fail */
685 if (!bd_may_claim(bdev, whole, holder))
688 /* if claiming is already in progress, wait for it to finish */
689 if (whole->bd_claiming) {
690 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
693 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
694 spin_unlock(&bdev_lock);
696 finish_wait(wq, &wait);
697 spin_lock(&bdev_lock);
706 * bd_start_claiming - start claiming a block device
707 * @bdev: block device of interest
708 * @holder: holder trying to claim @bdev
710 * @bdev is about to be opened exclusively. Check @bdev can be opened
711 * exclusively and mark that an exclusive open is in progress. Each
712 * successful call to this function must be matched with a call to
713 * either bd_finish_claiming() or bd_abort_claiming() (which do not
716 * This function is used to gain exclusive access to the block device
717 * without actually causing other exclusive open attempts to fail. It
718 * should be used when the open sequence itself requires exclusive
719 * access but may subsequently fail.
725 * Pointer to the block device containing @bdev on success, ERR_PTR()
728 static struct block_device *bd_start_claiming(struct block_device *bdev,
731 struct gendisk *disk;
732 struct block_device *whole;
738 * @bdev might not have been initialized properly yet, look up
739 * and grab the outer block device the hard way.
741 disk = get_gendisk(bdev->bd_dev, &partno);
743 return ERR_PTR(-ENXIO);
746 * Normally, @bdev should equal what's returned from bdget_disk()
747 * if partno is 0; however, some drivers (floppy) use multiple
748 * bdev's for the same physical device and @bdev may be one of the
749 * aliases. Keep @bdev if partno is 0. This means claimer
750 * tracking is broken for those devices but it has always been that
754 whole = bdget_disk(disk, 0);
756 whole = bdgrab(bdev);
758 module_put(disk->fops->owner);
761 return ERR_PTR(-ENOMEM);
763 /* prepare to claim, if successful, mark claiming in progress */
764 spin_lock(&bdev_lock);
766 err = bd_prepare_to_claim(bdev, whole, holder);
768 whole->bd_claiming = holder;
769 spin_unlock(&bdev_lock);
772 spin_unlock(&bdev_lock);
779 struct bd_holder_disk {
780 struct list_head list;
781 struct gendisk *disk;
785 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
786 struct gendisk *disk)
788 struct bd_holder_disk *holder;
790 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
791 if (holder->disk == disk)
796 static int add_symlink(struct kobject *from, struct kobject *to)
798 return sysfs_create_link(from, to, kobject_name(to));
801 static void del_symlink(struct kobject *from, struct kobject *to)
803 sysfs_remove_link(from, kobject_name(to));
807 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
808 * @bdev: the claimed slave bdev
809 * @disk: the holding disk
811 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
813 * This functions creates the following sysfs symlinks.
815 * - from "slaves" directory of the holder @disk to the claimed @bdev
816 * - from "holders" directory of the @bdev to the holder @disk
818 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
819 * passed to bd_link_disk_holder(), then:
821 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
822 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
824 * The caller must have claimed @bdev before calling this function and
825 * ensure that both @bdev and @disk are valid during the creation and
826 * lifetime of these symlinks.
832 * 0 on success, -errno on failure.
834 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
836 struct bd_holder_disk *holder;
839 mutex_lock(&bdev->bd_mutex);
841 WARN_ON_ONCE(!bdev->bd_holder);
843 /* FIXME: remove the following once add_disk() handles errors */
844 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
847 holder = bd_find_holder_disk(bdev, disk);
853 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
859 INIT_LIST_HEAD(&holder->list);
863 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
867 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
871 * bdev could be deleted beneath us which would implicitly destroy
872 * the holder directory. Hold on to it.
874 kobject_get(bdev->bd_part->holder_dir);
876 list_add(&holder->list, &bdev->bd_holder_disks);
880 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
884 mutex_unlock(&bdev->bd_mutex);
887 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
890 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
891 * @bdev: the calimed slave bdev
892 * @disk: the holding disk
894 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
899 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
901 struct bd_holder_disk *holder;
903 mutex_lock(&bdev->bd_mutex);
905 holder = bd_find_holder_disk(bdev, disk);
907 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
908 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
909 del_symlink(bdev->bd_part->holder_dir,
910 &disk_to_dev(disk)->kobj);
911 kobject_put(bdev->bd_part->holder_dir);
912 list_del_init(&holder->list);
916 mutex_unlock(&bdev->bd_mutex);
918 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
922 * flush_disk - invalidates all buffer-cache entries on a disk
924 * @bdev: struct block device to be flushed
925 * @kill_dirty: flag to guide handling of dirty inodes
927 * Invalidates all buffer-cache entries on a disk. It should be called
928 * when a disk has been changed -- either by a media change or online
931 static void flush_disk(struct block_device *bdev, bool kill_dirty)
933 if (__invalidate_device(bdev, kill_dirty)) {
934 char name[BDEVNAME_SIZE] = "";
937 disk_name(bdev->bd_disk, 0, name);
938 printk(KERN_WARNING "VFS: busy inodes on changed media or "
939 "resized disk %s\n", name);
944 if (disk_part_scan_enabled(bdev->bd_disk))
945 bdev->bd_invalidated = 1;
949 * check_disk_size_change - checks for disk size change and adjusts bdev size.
950 * @disk: struct gendisk to check
951 * @bdev: struct bdev to adjust.
953 * This routine checks to see if the bdev size does not match the disk size
954 * and adjusts it if it differs.
956 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
958 loff_t disk_size, bdev_size;
960 disk_size = (loff_t)get_capacity(disk) << 9;
961 bdev_size = i_size_read(bdev->bd_inode);
962 if (disk_size != bdev_size) {
963 char name[BDEVNAME_SIZE];
965 disk_name(disk, 0, name);
967 "%s: detected capacity change from %lld to %lld\n",
968 name, bdev_size, disk_size);
969 i_size_write(bdev->bd_inode, disk_size);
970 flush_disk(bdev, false);
973 EXPORT_SYMBOL(check_disk_size_change);
976 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
977 * @disk: struct gendisk to be revalidated
979 * This routine is a wrapper for lower-level driver's revalidate_disk
980 * call-backs. It is used to do common pre and post operations needed
981 * for all revalidate_disk operations.
983 int revalidate_disk(struct gendisk *disk)
985 struct block_device *bdev;
988 if (disk->fops->revalidate_disk)
989 ret = disk->fops->revalidate_disk(disk);
991 bdev = bdget_disk(disk, 0);
995 mutex_lock(&bdev->bd_mutex);
996 check_disk_size_change(disk, bdev);
997 bdev->bd_invalidated = 0;
998 mutex_unlock(&bdev->bd_mutex);
1002 EXPORT_SYMBOL(revalidate_disk);
1005 * This routine checks whether a removable media has been changed,
1006 * and invalidates all buffer-cache-entries in that case. This
1007 * is a relatively slow routine, so we have to try to minimize using
1008 * it. Thus it is called only upon a 'mount' or 'open'. This
1009 * is the best way of combining speed and utility, I think.
1010 * People changing diskettes in the middle of an operation deserve
1013 int check_disk_change(struct block_device *bdev)
1015 struct gendisk *disk = bdev->bd_disk;
1016 const struct block_device_operations *bdops = disk->fops;
1017 unsigned int events;
1019 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1020 DISK_EVENT_EJECT_REQUEST);
1021 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1024 flush_disk(bdev, true);
1025 if (bdops->revalidate_disk)
1026 bdops->revalidate_disk(bdev->bd_disk);
1030 EXPORT_SYMBOL(check_disk_change);
1032 void bd_set_size(struct block_device *bdev, loff_t size)
1034 unsigned bsize = bdev_logical_block_size(bdev);
1036 mutex_lock(&bdev->bd_inode->i_mutex);
1037 i_size_write(bdev->bd_inode, size);
1038 mutex_unlock(&bdev->bd_inode->i_mutex);
1039 while (bsize < PAGE_CACHE_SIZE) {
1044 bdev->bd_block_size = bsize;
1045 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1047 EXPORT_SYMBOL(bd_set_size);
1049 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1054 * mutex_lock(part->bd_mutex)
1055 * mutex_lock_nested(whole->bd_mutex, 1)
1058 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1060 struct gendisk *disk;
1061 struct module *owner;
1066 if (mode & FMODE_READ)
1068 if (mode & FMODE_WRITE)
1071 * hooks: /n/, see "layering violations".
1074 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1084 disk = get_gendisk(bdev->bd_dev, &partno);
1087 owner = disk->fops->owner;
1089 disk_block_events(disk);
1090 mutex_lock_nested(&bdev->bd_mutex, for_part);
1091 if (!bdev->bd_openers) {
1092 bdev->bd_disk = disk;
1093 bdev->bd_queue = disk->queue;
1094 bdev->bd_contains = bdev;
1096 struct backing_dev_info *bdi;
1099 bdev->bd_part = disk_get_part(disk, partno);
1104 if (disk->fops->open) {
1105 ret = disk->fops->open(bdev, mode);
1106 if (ret == -ERESTARTSYS) {
1107 /* Lost a race with 'disk' being
1108 * deleted, try again.
1111 disk_put_part(bdev->bd_part);
1112 bdev->bd_part = NULL;
1113 bdev->bd_disk = NULL;
1114 bdev->bd_queue = NULL;
1115 mutex_unlock(&bdev->bd_mutex);
1116 disk_unblock_events(disk);
1124 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1125 bdi = blk_get_backing_dev_info(bdev);
1127 bdi = &default_backing_dev_info;
1128 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1132 * If the device is invalidated, rescan partition
1133 * if open succeeded or failed with -ENOMEDIUM.
1134 * The latter is necessary to prevent ghost
1135 * partitions on a removed medium.
1137 if (bdev->bd_invalidated) {
1139 rescan_partitions(disk, bdev);
1140 else if (ret == -ENOMEDIUM)
1141 invalidate_partitions(disk, bdev);
1146 struct block_device *whole;
1147 whole = bdget_disk(disk, 0);
1152 ret = __blkdev_get(whole, mode, 1);
1155 bdev->bd_contains = whole;
1156 bdev_inode_switch_bdi(bdev->bd_inode,
1157 whole->bd_inode->i_data.backing_dev_info);
1158 bdev->bd_part = disk_get_part(disk, partno);
1159 if (!(disk->flags & GENHD_FL_UP) ||
1160 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1164 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1167 if (bdev->bd_contains == bdev) {
1169 if (bdev->bd_disk->fops->open)
1170 ret = bdev->bd_disk->fops->open(bdev, mode);
1171 /* the same as first opener case, read comment there */
1172 if (bdev->bd_invalidated) {
1174 rescan_partitions(bdev->bd_disk, bdev);
1175 else if (ret == -ENOMEDIUM)
1176 invalidate_partitions(bdev->bd_disk, bdev);
1179 goto out_unlock_bdev;
1181 /* only one opener holds refs to the module and disk */
1187 bdev->bd_part_count++;
1188 mutex_unlock(&bdev->bd_mutex);
1189 disk_unblock_events(disk);
1193 disk_put_part(bdev->bd_part);
1194 bdev->bd_disk = NULL;
1195 bdev->bd_part = NULL;
1196 bdev->bd_queue = NULL;
1197 bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1198 if (bdev != bdev->bd_contains)
1199 __blkdev_put(bdev->bd_contains, mode, 1);
1200 bdev->bd_contains = NULL;
1202 mutex_unlock(&bdev->bd_mutex);
1203 disk_unblock_events(disk);
1213 * blkdev_get - open a block device
1214 * @bdev: block_device to open
1215 * @mode: FMODE_* mask
1216 * @holder: exclusive holder identifier
1218 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1219 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1220 * @holder is invalid. Exclusive opens may nest for the same @holder.
1222 * On success, the reference count of @bdev is unchanged. On failure,
1229 * 0 on success, -errno on failure.
1231 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1233 struct block_device *whole = NULL;
1236 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1238 if ((mode & FMODE_EXCL) && holder) {
1239 whole = bd_start_claiming(bdev, holder);
1240 if (IS_ERR(whole)) {
1242 return PTR_ERR(whole);
1246 res = __blkdev_get(bdev, mode, 0);
1249 struct gendisk *disk = whole->bd_disk;
1251 /* finish claiming */
1252 mutex_lock(&bdev->bd_mutex);
1253 spin_lock(&bdev_lock);
1256 BUG_ON(!bd_may_claim(bdev, whole, holder));
1258 * Note that for a whole device bd_holders
1259 * will be incremented twice, and bd_holder
1260 * will be set to bd_may_claim before being
1263 whole->bd_holders++;
1264 whole->bd_holder = bd_may_claim;
1266 bdev->bd_holder = holder;
1269 /* tell others that we're done */
1270 BUG_ON(whole->bd_claiming != holder);
1271 whole->bd_claiming = NULL;
1272 wake_up_bit(&whole->bd_claiming, 0);
1274 spin_unlock(&bdev_lock);
1277 * Block event polling for write claims if requested. Any
1278 * write holder makes the write_holder state stick until
1279 * all are released. This is good enough and tracking
1280 * individual writeable reference is too fragile given the
1281 * way @mode is used in blkdev_get/put().
1283 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1284 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1285 bdev->bd_write_holder = true;
1286 disk_block_events(disk);
1289 mutex_unlock(&bdev->bd_mutex);
1295 EXPORT_SYMBOL(blkdev_get);
1298 * blkdev_get_by_path - open a block device by name
1299 * @path: path to the block device to open
1300 * @mode: FMODE_* mask
1301 * @holder: exclusive holder identifier
1303 * Open the blockdevice described by the device file at @path. @mode
1304 * and @holder are identical to blkdev_get().
1306 * On success, the returned block_device has reference count of one.
1312 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1314 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1317 struct block_device *bdev;
1320 bdev = lookup_bdev(path);
1324 err = blkdev_get(bdev, mode, holder);
1326 return ERR_PTR(err);
1328 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1329 blkdev_put(bdev, mode);
1330 return ERR_PTR(-EACCES);
1335 EXPORT_SYMBOL(blkdev_get_by_path);
1338 * blkdev_get_by_dev - open a block device by device number
1339 * @dev: device number of block device to open
1340 * @mode: FMODE_* mask
1341 * @holder: exclusive holder identifier
1343 * Open the blockdevice described by device number @dev. @mode and
1344 * @holder are identical to blkdev_get().
1346 * Use it ONLY if you really do not have anything better - i.e. when
1347 * you are behind a truly sucky interface and all you are given is a
1348 * device number. _Never_ to be used for internal purposes. If you
1349 * ever need it - reconsider your API.
1351 * On success, the returned block_device has reference count of one.
1357 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1359 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1361 struct block_device *bdev;
1366 return ERR_PTR(-ENOMEM);
1368 err = blkdev_get(bdev, mode, holder);
1370 return ERR_PTR(err);
1374 EXPORT_SYMBOL(blkdev_get_by_dev);
1376 static int blkdev_open(struct inode * inode, struct file * filp)
1378 struct block_device *bdev;
1381 * Preserve backwards compatibility and allow large file access
1382 * even if userspace doesn't ask for it explicitly. Some mkfs
1383 * binary needs it. We might want to drop this workaround
1384 * during an unstable branch.
1386 filp->f_flags |= O_LARGEFILE;
1388 if (filp->f_flags & O_NDELAY)
1389 filp->f_mode |= FMODE_NDELAY;
1390 if (filp->f_flags & O_EXCL)
1391 filp->f_mode |= FMODE_EXCL;
1392 if ((filp->f_flags & O_ACCMODE) == 3)
1393 filp->f_mode |= FMODE_WRITE_IOCTL;
1395 bdev = bd_acquire(inode);
1399 filp->f_mapping = bdev->bd_inode->i_mapping;
1401 return blkdev_get(bdev, filp->f_mode, filp);
1404 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1406 struct gendisk *disk = bdev->bd_disk;
1407 struct block_device *victim = NULL;
1409 mutex_lock_nested(&bdev->bd_mutex, for_part);
1411 bdev->bd_part_count--;
1413 if (!--bdev->bd_openers) {
1414 WARN_ON_ONCE(bdev->bd_holders);
1415 sync_blockdev(bdev);
1417 /* ->release can cause the old bdi to disappear,
1418 * so must switch it out first
1420 bdev_inode_switch_bdi(bdev->bd_inode,
1421 &default_backing_dev_info);
1423 if (bdev->bd_contains == bdev) {
1424 if (disk->fops->release)
1425 disk->fops->release(disk, mode);
1427 if (!bdev->bd_openers) {
1428 struct module *owner = disk->fops->owner;
1430 disk_put_part(bdev->bd_part);
1431 bdev->bd_part = NULL;
1432 bdev->bd_disk = NULL;
1433 if (bdev != bdev->bd_contains)
1434 victim = bdev->bd_contains;
1435 bdev->bd_contains = NULL;
1440 mutex_unlock(&bdev->bd_mutex);
1443 __blkdev_put(victim, mode, 1);
1446 void blkdev_put(struct block_device *bdev, fmode_t mode)
1448 mutex_lock(&bdev->bd_mutex);
1450 if (mode & FMODE_EXCL) {
1454 * Release a claim on the device. The holder fields
1455 * are protected with bdev_lock. bd_mutex is to
1456 * synchronize disk_holder unlinking.
1458 spin_lock(&bdev_lock);
1460 WARN_ON_ONCE(--bdev->bd_holders < 0);
1461 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1463 /* bd_contains might point to self, check in a separate step */
1464 if ((bdev_free = !bdev->bd_holders))
1465 bdev->bd_holder = NULL;
1466 if (!bdev->bd_contains->bd_holders)
1467 bdev->bd_contains->bd_holder = NULL;
1469 spin_unlock(&bdev_lock);
1472 * If this was the last claim, remove holder link and
1473 * unblock evpoll if it was a write holder.
1475 if (bdev_free && bdev->bd_write_holder) {
1476 disk_unblock_events(bdev->bd_disk);
1477 bdev->bd_write_holder = false;
1482 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1483 * event. This is to ensure detection of media removal commanded
1484 * from userland - e.g. eject(1).
1486 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1488 mutex_unlock(&bdev->bd_mutex);
1490 __blkdev_put(bdev, mode, 0);
1492 EXPORT_SYMBOL(blkdev_put);
1494 static int blkdev_close(struct inode * inode, struct file * filp)
1496 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1497 blkdev_put(bdev, filp->f_mode);
1501 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1503 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1504 fmode_t mode = file->f_mode;
1507 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1508 * to updated it before every ioctl.
1510 if (file->f_flags & O_NDELAY)
1511 mode |= FMODE_NDELAY;
1513 mode &= ~FMODE_NDELAY;
1515 return blkdev_ioctl(bdev, mode, cmd, arg);
1519 * Write data to the block device. Only intended for the block device itself
1520 * and the raw driver which basically is a fake block device.
1522 * Does not take i_mutex for the write and thus is not for general purpose
1525 ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1526 unsigned long nr_segs, loff_t pos)
1528 struct file *file = iocb->ki_filp;
1529 struct blk_plug plug;
1532 BUG_ON(iocb->ki_pos != pos);
1534 blk_start_plug(&plug);
1535 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1536 if (ret > 0 || ret == -EIOCBQUEUED) {
1539 err = generic_write_sync(file, pos, ret);
1540 if (err < 0 && ret > 0)
1543 blk_finish_plug(&plug);
1546 EXPORT_SYMBOL_GPL(blkdev_aio_write);
1548 static ssize_t blkdev_aio_read(struct kiocb *iocb, const struct iovec *iov,
1549 unsigned long nr_segs, loff_t pos)
1551 struct file *file = iocb->ki_filp;
1552 struct inode *bd_inode = file->f_mapping->host;
1553 loff_t size = i_size_read(bd_inode);
1559 if (size < iocb->ki_left)
1560 nr_segs = iov_shorten((struct iovec *)iov, nr_segs, size);
1561 return generic_file_aio_read(iocb, iov, nr_segs, pos);
1565 * Try to release a page associated with block device when the system
1566 * is under memory pressure.
1568 static int blkdev_releasepage(struct page *page, gfp_t wait)
1570 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1572 if (super && super->s_op->bdev_try_to_free_page)
1573 return super->s_op->bdev_try_to_free_page(super, page, wait);
1575 return try_to_free_buffers(page);
1578 static const struct address_space_operations def_blk_aops = {
1579 .readpage = blkdev_readpage,
1580 .writepage = blkdev_writepage,
1581 .write_begin = blkdev_write_begin,
1582 .write_end = blkdev_write_end,
1583 .writepages = generic_writepages,
1584 .releasepage = blkdev_releasepage,
1585 .direct_IO = blkdev_direct_IO,
1588 const struct file_operations def_blk_fops = {
1589 .open = blkdev_open,
1590 .release = blkdev_close,
1591 .llseek = block_llseek,
1592 .read = do_sync_read,
1593 .write = do_sync_write,
1594 .aio_read = blkdev_aio_read,
1595 .aio_write = blkdev_aio_write,
1596 .mmap = generic_file_mmap,
1597 .fsync = blkdev_fsync,
1598 .unlocked_ioctl = block_ioctl,
1599 #ifdef CONFIG_COMPAT
1600 .compat_ioctl = compat_blkdev_ioctl,
1602 .splice_read = generic_file_splice_read,
1603 .splice_write = generic_file_splice_write,
1606 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1609 mm_segment_t old_fs = get_fs();
1611 res = blkdev_ioctl(bdev, 0, cmd, arg);
1616 EXPORT_SYMBOL(ioctl_by_bdev);
1619 * lookup_bdev - lookup a struct block_device by name
1620 * @pathname: special file representing the block device
1622 * Get a reference to the blockdevice at @pathname in the current
1623 * namespace if possible and return it. Return ERR_PTR(error)
1626 struct block_device *lookup_bdev(const char *pathname)
1628 struct block_device *bdev;
1629 struct inode *inode;
1633 if (!pathname || !*pathname)
1634 return ERR_PTR(-EINVAL);
1636 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1638 return ERR_PTR(error);
1640 inode = path.dentry->d_inode;
1642 if (!S_ISBLK(inode->i_mode))
1645 if (path.mnt->mnt_flags & MNT_NODEV)
1648 bdev = bd_acquire(inode);
1655 bdev = ERR_PTR(error);
1658 EXPORT_SYMBOL(lookup_bdev);
1660 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1662 struct super_block *sb = get_super(bdev);
1667 * no need to lock the super, get_super holds the
1668 * read mutex so the filesystem cannot go away
1669 * under us (->put_super runs with the write lock
1672 shrink_dcache_sb(sb);
1673 res = invalidate_inodes(sb, kill_dirty);
1676 invalidate_bdev(bdev);
1679 EXPORT_SYMBOL(__invalidate_device);
1681 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1683 struct inode *inode, *old_inode = NULL;
1685 spin_lock(&inode_sb_list_lock);
1686 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1687 struct address_space *mapping = inode->i_mapping;
1689 spin_lock(&inode->i_lock);
1690 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1691 mapping->nrpages == 0) {
1692 spin_unlock(&inode->i_lock);
1696 spin_unlock(&inode->i_lock);
1697 spin_unlock(&inode_sb_list_lock);
1699 * We hold a reference to 'inode' so it couldn't have been
1700 * removed from s_inodes list while we dropped the
1701 * inode_sb_list_lock. We cannot iput the inode now as we can
1702 * be holding the last reference and we cannot iput it under
1703 * inode_sb_list_lock. So we keep the reference and iput it
1709 func(I_BDEV(inode), arg);
1711 spin_lock(&inode_sb_list_lock);
1713 spin_unlock(&inode_sb_list_lock);